Title: Distributed constraint satisfaction with multiply sectioned constraint networks
Authors: Yang Xiang; Younis Mohamed; Wanling Zhang
Addresses: School of Computer Science, College of Physical and Engineering Science, University of Guelph, 50 Stone Road East, Guelph, Ontario, N1G 2W1, Canada ' School of Computer Science, College of Physical and Engineering Science, University of Guelph, 50 Stone Road East, Guelph, Ontario, N1G 2W1, Canada ' School of Computer Science, College of Physical and Engineering Science, University of Guelph, 50 Stone Road East, Guelph, Ontario, N1G 2W1, Canada
Abstract: We propose a new algorithmic framework, multiply sectioned constraint networks (MSCNs), for solving distributed constraint satisfaction problems (DisCSPs) with complex local problems. An MSCN is converted into a linked junction forest (LJF) and is solved by a complete algorithm. Its time complexity is linear on the number and size of local problems (each in charge by an agent) and is exponential on cluster size of LJF. We show that the MSCN-LJF algorithm is more efficient than junction tree-based DisCSP algorithms. When a DisCSP is not naturally an MSCN, we show how to convert it into an MSCN, so that any DisCSP can be solved as above.
Keywords: distributed CSP; constraint satisfaction problems; multiply sectioned constraint networks; linked junction forests; LJF; complex local problems; multi-agent systems; agent-based systems; MAS; junction trees.
DOI: 10.1504/IJIDS.2014.061771
International Journal of Information and Decision Sciences, 2014 Vol.6 No.2, pp.127 - 152
Published online: 05 Jul 2014 *
Full-text access for editors Access for subscribers Purchase this article Comment on this article